Multiplayer coding with AI agents: one shared session for the team
What a shared agent session is, how live cursors and approvals work, and why it beats everyone running Claude or Codex alone.
A shared session is one workspace on one branch of code, joined by every teammate and every agent they run. Everyone sees the same files, cursors and agent output as it happens.
Because agents can see each other, the session becomes the coordination layer. Plans are visible before code is written; approvals are requested from whoever is present; a task can be handed to someone else with its full context.
Compared with each developer prompting alone in a terminal, a shared session removes the two most expensive failure modes: doing the same work twice and not knowing until review.